臉書 FaceBook 更新PyTorch，新版本支援評估和檢查模型的可視化工具
臉書 FaceBook 更新PyTorch，新版本支援評估和檢查模型的可視化工具
News from: iThome & FaceBook Developer Web Site.
臉書更新自家開源深度學習框架PyTorch，新增支援評估和檢查模型的可視化工具TensorBoard、改善即時編譯器等功能。
臉書在2019年度F8大會上宣布更新自家的開源深度學習框架PyTorch，1.1版新增了生產規模的開發工具，PyTorch在2017年首次發布，去年12月發布1.0版，這次的新版本則是新增支援評估和檢查模型的可視化工具TensorBoard、改善即時（JIT）編譯器、支援布林張量（Boolean tensors）和客製化的回歸神經網路，以及改善常見模型的效能表現，像是卷積式神經網路。
新版PyTorch針對TensorBoard的視覺化和模型除錯提供原生支援，TensorBoard為評估和檢查模型的可視化工具，針對即時編譯器的部分，臉書則是修復了錯誤和在TorchScript中擴展功能，像是支援dictionaries、user classes和attributes。
臉書也表示，現在正在用PyTorch 1.0版框架，來大規模地建置和部署機器翻譯、自然語言處理服務，這些系統每天提供像是在Messenger中的即時翻譯應用，將近60億次翻譯服務，且許多企業也透過PyTorch來加速新AI系統的開發和部署，像是Airbnb、ATOM、Genentech、微軟、Toyota研究院，而在大學教育方面，包含史丹佛大學的自然語言處理、加州大學柏克萊分校的電腦視覺和加州理工學院的機器人課程，都是透過PyTorch框架來教學。
------------------------------------------------------------------------------
News from: iThome & FaceBook Developer Web Site.
臉書更新自家開源深度學習框架PyTorch，新增支援評估和檢查模型的可視化工具TensorBoard、改善即時編譯器等功能。
臉書在2019年度F8大會上宣布更新自家的開源深度學習框架PyTorch，1.1版新增了生產規模的開發工具，PyTorch在2017年首次發布，去年12月發布1.0版，這次的新版本則是新增支援評估和檢查模型的可視化工具TensorBoard、改善即時（JIT）編譯器、支援布林張量（Boolean tensors）和客製化的回歸神經網路，以及改善常見模型的效能表現，像是卷積式神經網路。
新版PyTorch針對TensorBoard的視覺化和模型除錯提供原生支援，TensorBoard為評估和檢查模型的可視化工具，針對即時編譯器的部分，臉書則是修復了錯誤和在TorchScript中擴展功能，像是支援dictionaries、user classes和attributes。
臉書也表示，現在正在用PyTorch 1.0版框架，來大規模地建置和部署機器翻譯、自然語言處理服務，這些系統每天提供像是在Messenger中的即時翻譯應用，將近60億次翻譯服務，且許多企業也透過PyTorch來加速新AI系統的開發和部署，像是Airbnb、ATOM、Genentech、微軟、Toyota研究院，而在大學教育方面，包含史丹佛大學的自然語言處理、加州大學柏克萊分校的電腦視覺和加州理工學院的機器人課程，都是透過PyTorch框架來教學。
------------------------------------------------------------------------------
PyTorch Adds New Dev Tools As It Hits Production Scale
Building on the initial launch of PyTorch in 2017, Facebook partnered with the AI community to ship the stable release of PyTorch 1.0 last December. Key features of PyTorch v1.1 include:
- TensorBoard: First-class and native support for visualization and model debugging with TensorBoard, a web application suite for inspecting and understanding training runs and graphs.
- JIT compiler: Improvements to just-in-time (JIT) compilation. These include bug fixes and expanded capabilities in TorchScript, such as support for dictionaries, user classes, and attributes.
- New APIs: Support for Boolean tensors and better support for custom recurrent neural networks.
- Distributed Training: Improved performance for common models such as CNNs, and support for multi-device modules. See the latest tutorials here.
Using PyTorch across industries
Facebook is now using PyTorch 1.0 end-to-end workflows for building and deploying translation and natural language processing (NLP) services at scale. These systems provide nearly 6 billion translations a day for applications such as realtime translation in Messenger. Many leading businesses are also moving to PyTorch to accelerate development and deployment of new AI systems, including Airbnb, ATOM, Genentech, Microsoft and Toyota Research Institute (TRI).
Expanding educational resources and ecosystem
University-level classes — including Stanford NLP, UC Berkeley Computer Vision, and Caltech Robotics courses — are now being taught on PyTorch. In addition, today, we’re announcing our sponsorship of a new Udacity course that covers important concepts around building private and secure AI.
PyTorch 1.0 benefits from a growing ecosystem of tools, libraries and models to further support and accelerate AI development. Today we're expanding this ecosystem with:
- BoTorch (Bayesian Optimization for PyTorch): A package for optimizing hyper-parameters in the model, for training and inference. More details here.
- Ax (Adaptive Experimentation): A platform for iterative experimentation with different parameters.
- Google AI Platform Notebooks: Provides a higher-level platform for authoring, training and deploying models.
- Google Colab: Enables running a PyTorch notebook in Colab, making it easy to get started with PyTorch.
More details available here.
Introducing App Review Merging and Revamped Notifications
We are excited to announce the launch of one of the top developer requests — Instagram and Marketing API permissions can now be found in the main App Review tab and can be submitted along with other permissions, meaning only one submission is required. We are launching revamped jewel notifications in the app dashboard which are designed to be more organized and actionable, allowing developers to see all notifications related to personal or business-owned apps.
More details available here.
Open source at F8
At F8, we open-sourced tools and frameworks that simplify machine learning experimentation and optimization, speed up test execution times, and help solve memory-related performance issues. See the full list below, and click through to get started with our latest releases.
The Ax platform makes it easier for developers to optimize their products and infrastructure by leveraging recent advances provided by BoTorch. Start using Ax.
BoTorch is a library for for Bayesian optimization (BO) research, built on PyTorch. BoTorch that significantly boosts developer efficiency by combining a modular design and use of Monte Carlo-based acquisition functions with PyTorch's auto-differentiation feature. Start using BoTorch.
The iOS development bridge (idb) makes it easy to build complex workflows so that automation on iOS can be distributed amongst a fleet of machines. Start using idb.
Memscout is an analysis tool that scouts out allocator inefficiencies and provides insights into memory allocation patterns of the process, and then presents a breakdown of statistics that can be used to quickly diagnose memory-related performance issues. Start using Memscout.
Mvfst features include stream multiplexing as a transport feature, 0-RTT connection establishment, better loss recovery, security built in from the ground up, and flexible congestion control. Start using mvfst.
More details available here.
Advancing self-supervision, CV, NLP to keep our platforms safer
We use AI in a wide range of applications at Facebook today — and one of the most important is helping keep people safe on our platforms. In order to make all these systems more effective, we need to continue to improve our AI in two areas in particular: understanding content and working effectively with less labeled training data. Here we are highlighting how we’re improving the accuracy and efficiency of our content understanding systems and finding new ways to do more with less supervised learning.
More details available here.
Building Inclusive AI at Facebook
Teaching machines to be inclusive may sound unusual, but it's important in order to ensure that AI-powered devices work well for everyone, regardless of their skin tone and other attributes. AI models may perform differently for different people, in part because they’re trained with datasets created by people, and these datasets may contain limitations, flaws, or other issues. The inclusive AI process provides a set of working guidelines to help researchers and programmers design data sets, measure product performance, and test new systems through the lens of inclusivity.
More details available here.
Navigating Ethical Design in Tech
Responsible design means being intentional about decisions, being cognizant of the values that inform those decisions, and doing our best to anticipate the impact those decisions might have. And it means being careful to not define the potential impact of our products too narrowly, but take a broader view of the social and political contexts in which they operate. Some of the design challenges we are are addressing are combating misinformation and challenges of memorialization that works for different cultures and traditions.
More details available here.
We have more highlights on today's Facebook news and announcements in our Day 2 Facebook Newsroom Roundup. You can also check out the keynote and session videos on demand, and get updates on F8's big moments through our Facebook for Developers page, Twitter and Instagram. Be sure to follow the conversation with the #F82019 hashtag.
留言
張貼留言